ArchiTECHt Daily: The practical advice that could prevent deep learning disappointment

I was heartened on Tuesday to read a new report by research firm Altimeter Group that introduces read
ArchiTECHt Daily: The practical advice that could prevent deep learning disappointment
By ARCHITECHT • Issue #8
I was heartened on Tuesday to read a new report by research firm Altimeter Group that introduces readers to artificial intelligence (as in today’s technology, not the concept) and actually offers some practical advice on how to capitalize on it.  A lot of the talk about AI right now focuses on super-intelligent systems and machines besting humans in complex games (by the way, an AI system called Libratus just crushed a group of professional poker players at Texas Hold'em), but those discussions—while great for page views and certainly interesting—won’t be moving the world’s economic needle. At least not directly.
Rather, it’s mass adoption of AI by mainstream enterprises that’s really going to have a huge impact. And if the rate of adoption, and use cases, of precursor technologies such as cloud computing and big data are any indicator, most enterprises won’t be working on the cutting edge of AI any time soon. Instead, their early use cases will likely be very targeted, rather mundane and, if done correctly, not completely disappointing.
There are three points from the Altimeter Group report that stood out to me. I’ll summarize them:
  1. Data is everything: In order to train systems accurately, you need lots of good, labeled data. Start thinking now about the problems you want to solve, and what data you have (or need) to do that.
  2. Your company is not Google: Tackle low-hanging fruit and optimize existing processes using proven technologies. Don’t think you need an entire AI team or to incorporate it across the business.
  3. There’s more to life than chatbots and digital assistants: Voice-controlled and conversational interfaces have a place, but it’s not everyplace. Depending on the application, your best AI interface might be code with a numerical output.
The report also goes into privacy, governance, the competitive advantage of data, and some other carryover issues from the big data era. That’s understandable if, like me, you view AI as just an evolution of what Hadoop started a decade ago. And those issues certainly have not yet been resolved within most companies, or across broader industries or institutions such as the federal government.
TL;DR: It’s a good, easy to digest report. You should read it.

If you wondering about that human-computer poker tournament. Source: FiveThirtyEight
If you wondering about that human-computer poker tournament. Source: FiveThirtyEight
Around the web: Artificial intelligence
This is a good overview of deep learning by Aditya Singh of Foundation Capital, as well as some feasible near-term predictions for how AI will be democratized.  •  Share
A more global, and cautious, view of this CB Insights data from earlier this month. The overall trend toward more intelligent apps and advanced data analysis is real, even if the firm applies a liberal definition of AI.
The idea of a third-party version of Alexa, Siri or Cortana is compelling—and it will take some serious cash to go to battle against the platform players.
Around the web: Cloud and infrastructure
Lost data, significant downtime, non-existent backups. There are a lot of teachable lessons in here for SaaS startups, and probably a few cautionary ones for buyers.
Many technologies have come and gone without really threatening the ubiquity of Amazon S3 for new object storage workloads. But where there’s a will—and there is a will for portable/hybrid infrastructure—there’s a way.
Another attempt to eliminate data center cables and potentially boost efficiency. 10 Gbps won’t cut it with the big boys, though.
Backblaze has been sharing the ins and out (and specs) of its massive cloud backup infrastructure for years. Its annual analysis of drive performance found an HGST model with zero failures during 2016.
More from the CRE team at Google Cloud on proper monitoring.  Co-author (and CRE director Dave Rensin) dropped also some major knowledge on the ArchiTECHt Show podcast recently.
Tintri is working on a chatbot for making it dead simple to deploy new storage infrastructure (presumably in a private cloud setting). If it sounds too good to be true, check out GitHub’s Hubot.
Around the web: All things data
These numbers are really quite scary if they’re accurate. You have to wonder how community edition users are being helped.
Come for the fact that eBay runs an Oracle+MongoDB architecture, stay for the lessons in how it solved a database issue that was jamming up the network.
Speaking of databases, Cloudera’s open source Kudu data store appears to have gained serious traction since it was released. Now it’s officially ready for production.
And that’s just in the official repository. As data science is still taking off, Microsoft’s acquisition of Revolution Analytics looks pretty smart.
University of Chicago researchers put out an interesting study on internet saturation using a trillion data points, and in the process learned that global sleep patterns are converging.
Source: Revolution Analytics
Source: Revolution Analytics
Around the web: Software, in general
More non-profits should do something like this in order to step up innovation and efficiency. Of course, most non-profits don’t raise $24 million in a single weekend.
“It’s not about being open source, it’s about having a lower barrier to adoption and being free, and context.”— Kiyoto Tamura, Treasure Data. A good writeup of what seems like a good talk.
Did you enjoy this issue?
The most interesting news, analysis, blog posts and research in cloud computing, artificial intelligence and software engineering. Delivered daily to your inbox. Curated by Derrick Harris. Check out the Architecht site at
Carefully curated by ARCHITECHT with Revue. If you were forwarded this newsletter and you like it, you can subscribe here. If you don't want these updates anymore, please unsubscribe here.